A web helper that waits for your next step
Cling helps a person coordinate small actions across web pages. Its authors tested it with 12 people on web tasks, not in a school district.
Beyond PromptingRead story
Lead story
Worker voice needs two measures, not one
A proposed workplace study would track both worker ideas put to use and whether people feel heard. It has no results yet.
Source: arXiv · Oct 20, 2026Read original source
AI Agency
Read story
More top stories
A useful prompt is not permission to remember it
Promptile looks for helpful tactics in past AI chats. The digest asks users to approve what gets saved, where it applies, and how to remove it.
Beyond PromptingRead story
Featured
A second clinical check can miss the same thing
A preprint asks clinics to test whether doctors and AI miss the same cases. It proposes an audit, not a proven patient-safety fix.
AI AgencyRead story
Research & Science
Readable AI parts do not prove it is safe
Zvi Mowshowitz finds Goodfire’s work promising but questions what readable model parts prove. His essay is commentary, not an independent audit.
AI AgencyRead story
Research & Science
A faster-science forecast still needs a public test
Jack Clark argues that AI may speed up science and asks who will share the gains. His IMF lecture is a forecast, not proof of broad economic change.
AI AgencyRead story
Beyond Prompting
More from this edition
-
Agents & AutomationRead story
Count the human correction, not just the finish
A preprint studies what people add during work with AI. The digest suggests recording which human correction changed the next step.